feat(voice): voice picker in settings — select from account voices including clones

[sterlingchad]

Problem

The voice assistant currently uses a hardcoded list of ElevenLabs pre-made voices with no way to select between them in the UI. Users who:

• Have an ElevenLabs paid plan and have created instant voice clones cannot use those clones with Hapi
• Want a different pre-made voice cannot change it without editing config manually

Proposed solution

Add a voice picker to Settings → Voice Assistant that:

1. Fetches the user's available voices dynamically from ElevenLabs (GET /v1/voices) via a hub proxy route, so cloned voices appear alongside pre-made voices
2. Shows a ▶ preview button per voice using the ElevenLabs preview_url (CDN-hosted, no auth needed to play)
3. Labels cloned voices distinctly so users can identify them
4. Persists the selection to localStorage and passes it as an override when starting a voice session (ElevenLabs already supports platform_settings.overrides.conversation_config_override.tts.voice_id)
5. Degrades gracefully — if no API key is configured, shows a static fallback list

Notes

• ElevenLabs voice override is already supported via overrides.tts.voiceId in startSession
• The hub proxy approach keeps the API key server-side (consistent with the existing token endpoint pattern)
• This lays the groundwork for a future phase where the voice provider can be abstracted (local TTS, Gemini voice, etc.)

[github-actions]

This is not implemented in the current codebase. Verified current behavior: Settings -> Voice Assistant only exposes voice language selection, and voice sessions only pass a language override to ElevenLabs. I did not find a hub route that proxies GET /v1/voices, a voice picker UI, preview playback, or persisted voice_id selection.

One correction from the notes: I did not find current tts.voice_id runtime override support. The generated ElevenLabs agent config hardcodes the default TTS voice, and the enabled platform override currently covers agent.language only.

Relevant code:

• web/src/routes/settings/index.tsx:754 - Voice Assistant settings section.
• web/src/routes/settings/index.tsx:767 - current setting is Voice Language, not voice identity.
• web/src/lib/voice-context.tsx:43 - reads hapi-voice-lang from localStorage when starting voice.
• web/src/realtime/RealtimeVoiceSession.tsx:76 - sends overrides.agent.language to ElevenLabs.
• shared/src/voice.ts:239 - default TTS voice_id is hardcoded to Jessica.
• shared/src/voice.ts:245 - runtime override config is currently for language selection.
• hub/src/web/routes/voice.ts:120 - existing hub voice API exposes token creation only.
• web/src/api/client.ts:528 - web client has fetchVoiceToken; no voices-list client method found.

---

HAPI Bot

[heavygee]

Related Discussion (roadmap/context): #691

[heavygee]

Closing as addressed by #690 (merged 2026-05-27), which added the dynamic ElevenLabs voice picker in Settings — fetching account voices including instant clones, with a preview button on each. Per-backend voice selection for Gemini Live and Qwen Realtime is tracked in #742 / PR #743.